Replies to post #55519 on Intel Corp (INTC)

[wbmw]

Chipgeek, I agree. Quad core with turbo mode is like having the best of both words. Great multithreaded performance through multiple cores, and great single threaded performance through higher frequencies.

Just look at the problem with Phenom. AMD is going from a 3.2GHz K8 dual core architecture to a 2.3GHz K8L quad core architecture, and in single threaded apps, it's a big step backwards.

Kentsfield and Yorkfield don't have as much of a problem, primarily because Intel has positioned high end quad core frequencies somewhat close to their top dual core frequencies. Of course, this is mostly artificial, since it ought to be possible to manufacture 3.x GHz dual cores, if quads are also hitting 3.0GHz. But as you mention, doing this creates a positioning issue and forces the consumer to chose between multithreaded performance or single threaded performance.

But with Nehalem, consumers won't have this problem. It's probably one of the leading features of the micro-architecture, and if it works well, Intel should put the spotlight directly on it.

[smooth2o]

re: does anything special in emphasizing Nehalem's Turbo Mode

I'm sure they will, but only Intel can figure out what to do.... One thing you could be concerned about is QC on the desktop and notebook (due in Q109) will do to the replacement rate... As QC software improves, who needs a new computer?

Smooth

[Saturn V]

Hi ChipGeek,
Multithreading and Turbo Boost will give rise to interesting issues for an Operating System on how to allocate threads between multiple cores.

For example if have 4 threads which need to be executed on a 4 core , 8 thread chip. How should the threads be allocated ?

To take advantage of Turbo Boost the four threads should be allocated to two cores, and this will allow taking advantage of TurboBoost. However in some situations having one thread per core,gives each thread exclusive use of the L1 cache and all the other ALU and other Core resources. I can see that if two threads belong to the same program, they would share data and thus a common core for both would make sense. However threads belonging to independent modules would not share data, and would benefit by having an exclusive access to the core resources.

So the thread allocation strategy is not obvious at this point in time.

Is turning on the Turbo Boost under OS control or is it hardwired into the hardware logic of the processor ?

Another issue is that if you have eight threads being executed, and then a few terminate, will it make sense for the OS to move the threads to get an optimal performance configuration ?

Interesting issues !

I agree that the TurboBoost will 'Boost' the sales of 4 core Nehalems on the desktop.